On 2008-10-08, Stuart Henderson  wrote:

quoted text
> On 2008-10-08, BARDOU Pierre  wrote:

>> This is a multi-part message in MIME format.

>>

>> ------=_NextPart_000_00C3_01C92936.6DEF4560

>> Content-Type: multipart/mixed;

>> 	boundary="----=_NextPart_001_00C4_01C92936.6DEF4560"

>>

>>

>> ------=_NextPart_001_00C4_01C92936.6DEF4560

>> Content-Type: text/plain;

>> 	charset="iso-8859-1"

>> Content-Transfer-Encoding: quoted-printable

>

> Ugh, I thought the list server stripped these. Did something change?

>

>> The problem is that if the ISP router fails, my corresponding BGP=20

>> router is still up and running, and so keeps the CARP master,=20

>> which makes him a black hole :(

>

> I don't think I'd do it like this (either preferring OSPF running

> on BGP speakers to distribute default routes, or iBGP to avoid

> handing traffic to one router only to hand it straight to the other

> one). But it can be done, look at "demote".

>

>

Oh, in case it wasn't clear, you also need to write the bgpd.conf

parts to handle route selection. As Claudio says, just the standard

traffic engineering methods. Investigate localpref, prepend-neighbor,

weights, etc. There is no magic "balance my traffic" button.

See http://quigon.bsws.de/papers/epf2006/mgp00012.html.
As you hopefully know, balacing incoming traffic is a different

matter. Return packets do not automatically come in via the ISP

where you sent the associated outbound packets. For this, look

at prepends and whether your upstreams give you any finer

control over traffic-engineering via communities (for an

example of what some providers let you do, see e.g. "whois

-r as3356", in the "Communities accepted from customers" section).
If you are learning this whole area, you have some reading to

do. Plenty of information is available online and in print.

Much of it is aimed at cisco users and you'll need to read

between the lines for any !cisco, but the basic information

and techniques are generally applicable.